| dc.description.abstract | The methodological approach of this doctoral thesis is to use optical imaging and to employ the mathematical theory of regularisation methods, in order to support cultural heritage research. State-of-the-art optical imaging systems allow new possibilities for the acquisition, analysis, and visualisation of tangible cultural heritage objects. The optical acquisitions are noninvasive and constitute a precise and persistent description of valuable and indeed fragile cultural heritage artefacts. The optical documentation of cultural heritage objects generally focuses on two main traits: an accurate colour acquisition and 3D structure recording of the surface. Both aspects are addressed in this doctoral thesis. An accurate colour acquisition requires going beyond conventional RGB imaging. Multiand hyperspectral imaging do so by sampling the spectral information more densely. A persistent record of a surface aims for the spectral reflectance that constitutes the measurement without the influence of the illumination and device. Such high-quality records allow for example the comparison with subsequent acquisitions and can be an indication of any changes that have occurred over time monitoring. For multispectral imaging, an improved method for spectral reflectance recovery, based on support vector regression, is presented in this thesis. Hyperspectral push-broom scanners capture a larger number of spectral bands compared to filter-based multispectral cameras. However, all optical acquisitions are inevitably corrupted by noise. To further improve the quality of the records, mathematical models for hyperspectral denoising are investigated. We propose a variational denoising approach as a post-processing. Optical imaging devices include additional sensor data, such as the measured quantum efficiency, or the dark current during the acquisition. We exploit this additional sensor knowledge and transform the acquired image into a photon-corrected image format that facilitates the noise estimation. A main contribution of this work is the development of a hyperspectral denoising framework that takes into account the hyperspectral cube structure, Poisson distributed noise and additional sensor knowledge. As a complete application of accurate spectral recordings allow data analysis and visualisation applications in cultural heritage research, we applied such analyses to Edvard Munch’s famous masterpiece, the 1893 version of The Scream. This research was conducted in close cooperation with the curators from the museum and the manufacturer of a hyperspectral pushbroom scanner. 3D information is another important characteristic in cultural heritage documentation. We investigate Neolithic stone engravings, and the most relevant information of the engravings are constituted in the surface structure. We acquired several datasets with a colour-corrected, structured light scanner, and use it as a foundation for an in-depth, contact-less examination. We extracted features from the 3D mesh that we statistically analysed in order to confront them with archaeological questions and hypotheses. Conducted work involves acquisition, analysis and visualisation of cultural heritage objects. A focus of this thesis is the mathematical optimisation of higher dimensional datasets and statistical evaluations. The methodological approach also involves a close interdisciplinary cooperation with cultural heritage experts. | nb_NO |
